Opened 10 years ago

Closed 8 years ago

#628 closed defect (fixed)

IZ2898: ACCT_RESERVED_USAGE ignores slot count

Reported by: rdickson Owned by:
Priority: normal Milestone:
Component: sge Version: 6.1u5
Severity: minor Keywords: man
Cc:

Description

[Imported from gridengine issuezilla http://gridengine.sunsource.net/issues/show_bug.cgi?id=2898]

        Issue #:      2898             Platform:     All      Reporter: rdickson (rdickson)
       Component:     gridengine          OS:        All
     Subcomponent:    man              Version:      6.1u5       CC:    None defined
        Status:       REOPENED         Priority:     P3
      Resolution:                     Issue type:    DEFECT
                                   Target milestone: ---
      Assigned to:    pollinger (pollinger)
      QA Contact:     andreas
          URL:
       * Summary:     ACCT_RESERVED_USAGE ignores slot count
   Status whiteboard:
      Attachments:

     Issue 2898 blocks:
   Votes for issue 2898:


   Opened: Thu Jan 29 14:33:00 -0700 2009 
------------------------


The execd_param ACCT_RESERVED_USAGE (and probably SHARETREE_RESERVED_USAGE too)
does not handle parallel environments in a reasonable way.  For example:

Jobs 231 is just "sleep 60" in a parallel environment.  ACCT_RESERVED_USAGE was
turned off while it ran.

> qacct -j 231 | egrep "wallclock|slots|cpu"
slots        4
ru_wallclock 60
cpu          0

We turn on ACCT_RESERVED_USAGE and run it again, job 233.  Its accounting record
looks like this:

 > qacct -j 233 | egrep "wallclock|slots|cpu"
slots        4
ru_wallclock 60
cpu          60

Seems to me the user ought to be charged for 240 cpu seconds --- 60 seconds on
each of 4 slots.

   ------- Additional comments from crei Thu Feb 26 02:09:51 -0700 2009 -------
When using parallel environment the submitter have to specify the nr. of slots
the job will use (qsub -pe):
qsub -pe mytestpe 1 $SGE_ROOT/examples/jobs/sleeper.sh 60
Your job 260736 ("Sleeper") has been submitted

> qacct -j 260736 | egrep "wallclock|slots|cpu"
slots        1
ru_wallclock 60
cpu          0

The man page of ACCT_RESERVED_USAGE does not mention to have an influence at the
slots value:

 ACCT_RESERVED_USAGE
          If this  parameter  is  set  to  true,  the   usage  of
          reserved  resources  is used for the accounting entries
          cpu, mem and io instead of the measured usage.

Therefore I close this issue and set it to invalid.


   ------- Additional comments from rdickson Fri Feb 27 06:59:44 -0700 2009 -------
Ok, then I'm going to reopen this issue against the man page.

The man entry, "the usage of reserved resources is used for the accounting entries cpu, mem and io instead of the measured usage," is
uninformative.  Does ACCT_RESERVED_USAGE have anything to do with resource reservation? (qsub -R y)  No, it does not.  Does it have anything
to do with reserved run time?  (qsub -l h_rt=hh:mm:ss)?  No, it does not, because the accounting entry made is not the *reserved* h_rt, it
is the *measured* wallclock time.  The name implies some entity is being reserved, and slots seemed the obvious thing.  But apparently I was
wrong.

Perhaps the feature has a misleading name?  I understand it would be futile to suggest renaming it.  But please change the man page to
describe what it does, and perhaps even what it's good for?

Change History (1)

comment:1 Changed 8 years ago by dlove

  • Resolution set to fixed
  • Severity set to minor
  • Status changed from new to closed

Fixed by [3915/sge]

Note: See TracTickets for help on using tickets.